Statistical model and estimation method
Multiple regression is most common statistical model
Estimation methods are typically ordinary least squares (OLS), OLS with adjustment for autocorrelation (i.e., variance correction and use of effective degrees-of-freedom), or generalized least squares (i.e., OLS after whitening)
Block/epoch-based or event-related model
Hemodynamic response function (HRF)
Assumed HRF model (e.g., SPM's canonical difference of gammas HRF; FSL's canonical gamma HRF), HRF basis (list basis set) or estimated HRF (supply methods for estimating HRF)?
Additional regressors used (e.g., temporal derivatives, motion, behavioral covariates)
Any orthogonalization of regressors
Drift modeling/high-pass filtering (e.g., “DCT with cut off of X seconds”; “Gaussian-weighted running line smoother, cut-off 100 seconds”, or “cubic polynomial”)
Autocorrelation model type (e.g., AR(1), AR(1) + WN, or arbitrary autocorrelation function), and whether global or local.
(e.g., for SPM2/SPM5, ‘Approximate AR(1) autocorrelation model estimated at omnibus F-significant voxels (P < 0.001), used globally over the whole brain’; for FSL, ‘Autocorrelation function estimated locally at each voxel, tapered and regularized in space.’).
Contrast construction
Exactly what terms are subtracted from what? Define these in terms of task or stimulus conditions (e.g., using abstract names such as AUDSTIM, VISSTIM) instead of underlying psychological concepts